Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
MP 7,1 Waiting for Mac Pro 7,1 A1991 (no more)
- Thread starter Mago
- Start date
-
- Tags
- macpro 2016
- Sort by reaction score
You’ve missed my point.
We know what the Intel processors will do, it’s been proven.
What I’m saying to you, is that for the same physical size we do not know what the ARM will do once scaled up. All we have is educated guesses and Apple, and your, marketing speak.
I could put an industrial CRAC unit on the back of my ARM Mac, (that might provide the same TFLOPS as my Intel version), but then I’ve moved away from an Apples to Apples comparison.
Designs, and materials, tend to behave differently, depending on scale. With architecture, materials can only build structures up to a certain size, before the weight of the necessary thicknesses of materials overtakes the structural strength of the materials. Or in rocketry - there are absolute limits on payloads, because eventually the weight of the fuel to power the lift engine will become heavier than the engine's lift capacity.
We don't know if the A(x), or ARM in general, is capable of working competitively when you scale it up. As you make any complex system larger, the support infrastructure becomes more complex, and you get into a race against declining returns. I'm simply keeping an open mind to the possibility of something inherent to ARM or the A(x) precluding the ability to just scale up to Xeon-level performance.
I actually have no idea what your point is.
Are you saying that the CPU architecture is going to make a 100 or 150W ARM chip impossible to cool with a traditional heat sink and fan cooling solution? Ampere’s 125W TDP ARM chip might disagree.
I have no idea why you’re talking about an industrial scale A/C system; you’re not trying to remove hundreds of thousands or millions of BTUs.
Cooling a 150W processor to 70-80° C is simply not that difficult. I honestly have no idea where you’re coming from.
Still far behind Intel/AMD, it may catch some low power i5 from 2 years ago but never current generation, not to say HEDT which is by own right another league.
Here you make a case, yes some users in the low end (those not clear about to buy a MBA or an iPad) may live well with slower chips as long web, YouTube email works fine, and those are very smp optimized applications, but then now Intel has cheap low-end it's that do not require a custom macOS (very expensive to develop) and do not require to apple to develop another chip for a market which is a fraction of the iPad s (most iPad do recycle iPhone Ax soc few has custom Ax soc ), this is something windows even can't afford, and the few ARM-win10 machines use celular-less versión if the latest and most powerful Qualcomm snapdragon soc, read about the reviews of these systems...
You said iPad, it means iOS, by long time iOS have been being optimized for smp, from GUI to storage, not the same thing as macOS where even some cad applications rely on single threaded libraries not yet possible to convert to multi-threaded, or very expensive to migrate to smp models.
I think the engineers who are making ARM server chips know more than you think they do.
Let’s also acknowledge that during the course of this convo the goalposts have been moved from an A13 of several watts being able to run without throttling when properly cooled, to scaling up to Xeon level performance.
Last edited:
There are few things not discussed, about arm chips as memory interface bandwidth, PCIE lines available, adding those like in a desktop are not trivial design tasks.
[automerge]1569507220[/automerge]
[automerge]1569507220[/automerge]
See where are used chips like calvium thunderX...
Never mind. The bottom line is this;
If Apple could have made this work and reduced their reliance on Intel, they would have done. Even if it meant stacking loads of cores/processors together.
Something fundamental is lacking on the performance side.
When it comes to cooling, I have far more faith in Apple being able to keep an A-Series CPU/GPU cool then an Intel or AMD one. I believe Apple's cooling problems are predominately due to Intel and AMD so badly missing their TDP targets they gave Apple (and other OEMs), who designed their chassis around chips that were expected to run a fair bit cooler then they did.
We have just as much cooling issues with the thin Dell and Lenovo laptops as we do the thin MacBook Pros so it's clearly not just an "Apple thing". Hell, our Dell mobile workstations which are as thick as a Steven King novel run hot with Xeons and upper-end nVidia GPUs in them.
We have just as much cooling issues with the thin Dell and Lenovo laptops as we do the thin MacBook Pros so it's clearly not just an "Apple thing". Hell, our Dell mobile workstations which are as thick as a Steven King novel run hot with Xeons and upper-end nVidia GPUs in them.
ARM servers are not packing single CPUs that excel at the wide variety of desktop productivity tasks.
ARM's performance, and Apple's implementation, is impressive when compared against Intel's inability to scale the X64 architecture down to the power / performance levels ARM works within, but you can't just assume it's fait accompli that Apple will be able to scale up into the power / performance realms where Intel (or AMD) is at its best.
For a start, I don't think Apple would bother using their crack mobile processor design resources to make a desktop CPU, when they're a commodity item from two competing suppliers, and Apple's competitive distinction in that market is software.
There are those who believe Apple will be transitioning MacOS to ARM soon enough. I don’t think there’s necessarily anything inherently flawed with ARM that will keep it from becoming increasingly more performant. The ARM instruction set has become pretty CISCy. Transistors are ultra cheap. Maybe Mac Pro and iMac Pro stay Intel longer, 16” MBP too but others go ARM within the next few years. Maybe the high end goes ARM as well but 3-5 years out... who knows? Not me, anyway.
I don’t disagree with most of what you’re saying, but keep in mind those high end processors are a very expensive commodity. But the volumes at the higher end could very well be too small for Apple to bother. 80% of Macs are laptops, however.
It's not so much about "missing" TDP targets as Intel's means of measuring TDP doesn't match how their CPUs work in a post Turbo Boost world. When your CPU has a feature to self-overclock, and your TDP is measured with it disabled or not active, it's a less useful metric when designing systems. I wonder if Intel realized that consumption under boost is varied enough that it doesn't exactly make sense to report a single TDP when each copy of the silicon will behave slightly differently under boost.
But at the same time, it's not like Apple and other OEMs aren't aware of this behavior. They have been seeing it in their engineering labs for years at this point. And Apple could configure their systems to obey the TDP when boosting, at the cost of folks complaining, if they really wanted to.
That said, I am pretty impressed that the 2018 Mac Mini is able to handle dissipating ~80-90W in the same case as the 2014, which used chips with less than half the TDP. And still be quiet while doing it.
Not necessarily. As others have pointed out, things like PCIe would need to be supported, and generally you need a lot better I/O support. Something that an AX SOC doesn't care about in iPhones or iPads. If you want to support Thunderbolt, you need PCIe lanes available. So even a laptop would need access to extra I/O if you wanted to hook up a Titan Ridge controller. So there's some "despecialization" that has to happen before an ARM chip could replace Intel/AMD in laptops or desktops.
Sometimes it isn't even about scaling the design, but the engineering resources required to do it, and how far to carry prototyping/etc. Apple is in a different world today than 2005. AMD and Intel both have a market they are interested in serving, and Apple is part of that market. It's not like when Motorola was wondering if they should care about non-embedded CPUs, and IBM wondering if they should care about non-server CPUs. But Apple's also different in that in 2005, Mac was their primary business by revenue with iPod a close second. Now, the iPhone is, in a market where CPUs is heavily driven by Qualcomm and Samsung. So I'd say the business case for custom SOCs is stronger for the iPhone than it is for the Mac.
I'd be surprised if Apple didn't have prototypes. iOS and macOS are already similar enough that you could do it and get a rough idea what the experience is like, by standing up WindowServer and AppKit on the iPad Pro. I'd also be surprised if Apple didn't squeeze the Intel and AMD relationships as much as possible before making the leap themselves in the coming years, to avoid a ton of investment in what is effectively commodity hardware (yet important commodity hardware to Apple because of macOS).
iOS GUI and storage is still single thread. It uses the same libraries as Mac (CoreAnimation is used for draw on both platforms now.)
There's no magical multithreaded draw library on iOS. It's all the same. Which is something the new Catalyst/iPad Apps on Mac use to their advantage. They can use the same system draw libraries.
There's a lot of storage libraries, but Apple's CoreData on both platforms is also the same. Same threading, same handling.
iOS is basically macOS, just with different UI widgets. All the underlying kernel/libraries are the same. The UI widget library is different, but the underlying draw stack is the same. Even places where the kernel is configured different on iOS (like no virtual memory), that's just a switch in the same shared kernel that could be flipped on the Mac.
Last edited:
Why not do a mixed product line? That seems to answer both sets of questions - ARM almost certainly does scale to small laptop/low end Mac Mini type performance (that's not very different from an iPad Pro). If ARM has power/thermal/cost advantages on the low end, it makes sense to use it.
Why does that mean that the 15" MBP, the bigger iMacs and the Mac Pro need to move? The App Store can handle the multiple binary issue easily enough (and ARM Macs could be restricted to the App Store, while Intel Macs could continue to get software anywhere)...
Why does that mean that the 15" MBP, the bigger iMacs and the Mac Pro need to move? The App Store can handle the multiple binary issue easily enough (and ARM Macs could be restricted to the App Store, while Intel Macs could continue to get software anywhere)...
I’ll eat my hat if they take a permanently split approach. Who wants to play “will this app run on both my Macs” game because one is arm one is Intel.
I hardly have the energy anymore to explain, yet again, that any claim that one ISA is somehow magically able to scale to desktop while another cannot should be treated with extreme distrust.
As you note, if you give me a big enough thermal envelope, I can design an ARM that competes with anything Intel has on the desktop.
My only concern in Apple’s case would be related to the fact that they seem to use more die area than should be necessary - I have theories as to why (see Intrinsity patents), but just theories, and am not sure whether that particular design style would be a good idea for a desktop-style processor. It would work, but may raise costs unnecessarily. In any event, I have no doubt that Apple could design a desktop processor if it wanted to. I also believe they have already done this.
Everyone else included Microsoft is taking a split approach for now. It may not last forever (who knows where Apple will be in a decade), but it seems do-able.
When Apple switched to Intel, Intel had a full lineup of CPUs already to go for every Apple product. That put Apple in a position where they didn't need to split the lineup. ARM won't be the same way.
It's not the first time this has happened either. 68k and PowerPC co-existed for a while.
Who is “everyone else” apart from Microsoft? Also note: the exact problem I described exists with a Windows on arm machine, does it not?
In Windows RT there were no existing apps available, and in Windows 10 on Arm you have to emulate everything (and no 64bit) and get performance worse than the lowest inte CPU’s released years earlier.
Apple has taken this approach before yes, as a transitional tool. It’s never been a long-term thing.
Windows 10 is JIT, not emulation (or at least traditional emulation.) JIT should be able to get native, or near native performance. And so far that seems to be true from the initial Windows 10 ARM laptops.
In addition to that, Windows 10 apps are being updated to ARM. That'll take a bit because Microsoft has a bigger ship to steer.
Last edited:
Microsoft taking their 90% of the market and splitting that 80:20 with x86 and ARM would leave their ARM "subsset" with a larger share than the Mac market in total ( 0.9 * 0.2 = 0.18 ... 18% more than double 7-8% ).
In contrast, Apple taking a 75:25 split with perhaps assigning that ARM:x86 would leave the x86 remnant at just 2% ( if start at 8% ). Which would put that down in the range of what Linux is ( or less given the folks they'd drive off ).
Apple could do it, but they'd be taking on more risk. Especially when get down the the low level driver support level where things may/may not be as simple as a recompile.
If take the fraction of number of systems ported divided by number of months to complete the total transition I suspect there are relatively close. For better or worse there was more stuff to port in 94-96 timeline than in the x86 one. Even more so if the reports that Apple farmed out the Mac Pro port primarily to Intel (and take that Mac Pro off the number of systems for PPC->x86 in that ratio).
PPC was ready for the transition. It was the then macOS that had substantive issues. ( bits that were crufty 68K code ). And also price point issues ( because Apple was still following a 'buy marketshare' like strategy ).
There's a few things to read into there.
First, Kexts are basically dead in 10.15. That's likely in advance of ARM, and on ARM I expect third party kexts will be mostly or completely dead. DriverKit looks like something that would provide for very quick porting between Intel and ARM. It looks a lot like Apple cleaning house ahead of ARM. And I've heard whispers that the Kext architecture will eventually be completely sealed off from third parties.
Second, iOS and Mac have the same driver architecture. That means any existing drivers (mostly for Apple's own chipsets and GPUs) will be ready for the Mac with little to no work.
For Apple's own hardware, they look to be ready to go for a Mac transition. For third party hardware, the answer seems to be Apple shoving everyone onto DriverKit.
It’s “transparent” binary emulation, exactly like Rosetta was. And rather than assuming that somehow it won’t be crap like it was then, you could just go look at the benchmarks people did and see for yourself that it is indeed crap.
If Mac does go ARM, the Mac Pro will be the very last product to make the transition (a petite MacBook likely the first), so we’ll be well into the Waiting for Mac Pro 9,1 thread by then.
Waiting... so much waiting.
This tariff-backed Made in USA nonsense has quite probably killed any hopes for a smooth launch for this Mac Pro. I doubt it’ll be like the dream rollout of the iMac Pro, where you ordered, you got.
Despite this thread’s naysayers, there is enormous pent-up demand for this product and I very much doubt the Austin facility will be able to produce enough units to ship them in a timely fashion.
I have the sense of December 2013 déjà vu - within the first few hours, orders were backed out to February, and months long waits were the norm for the next half year.
Waiting... so much waiting.
This tariff-backed Made in USA nonsense has quite probably killed any hopes for a smooth launch for this Mac Pro. I doubt it’ll be like the dream rollout of the iMac Pro, where you ordered, you got.
Despite this thread’s naysayers, there is enormous pent-up demand for this product and I very much doubt the Austin facility will be able to produce enough units to ship them in a timely fashion.
I have the sense of December 2013 déjà vu - within the first few hours, orders were backed out to February, and months long waits were the norm for the next half year.
Yet another ****** thing done by that orange moron